Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[SD3 Inference] T5 Token limit #8506

Merged
merged 7 commits into from
Jun 18, 2024
Merged

[SD3 Inference] T5 Token limit #8506

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
6381c1a
max_sequence_length for the T5
asomoza Jun 13, 2024
5824176
updated img2img
asomoza Jun 13, 2024
5e553a7
Merge branch 'main' into token-max-length
asomoza Jun 13, 2024
553426e
Merge branch 'main' into token-max-length
sayakpaul Jun 16, 2024
90416ee
apply suggestions
asomoza Jun 17, 2024
54ddf68
Merge branch 'main' into token-max-length
asomoza Jun 17, 2024
1cd4333
Merge branch 'main' into token-max-length
yiyixuxu Jun 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 18 additions & 4 deletions src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,6 +205,7 @@ def _get_t5_prompt_embeds(
self,
prompt: Union[str, List[str]] = None,
num_images_per_prompt: int = 1,
max_sequence_length: int = 256,
asomoza marked this conversation as resolved.
Show resolved Hide resolved
device: Optional[torch.device] = None,
dtype: Optional[torch.dtype] = None,
):
Expand All @@ -224,7 +225,7 @@ def _get_t5_prompt_embeds(
text_inputs = self.tokenizer_3(
prompt,
padding="max_length",
max_length=self.tokenizer_max_length,
asomoza marked this conversation as resolved.
Show resolved Hide resolved
max_length=max_sequence_length,
truncation=True,
add_special_tokens=True,
return_tensors="pt",
Expand All @@ -235,8 +236,8 @@ def _get_t5_prompt_embeds(
if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
removed_text = self.tokenizer_3.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
logger.warning(
"The following part of your input was truncated because CLIP can only handle sequences up to"
f" {self.tokenizer_max_length} tokens: {removed_text}"
"The following part of your input was truncated because `max_sequence_length` is set to "
f" {max_sequence_length} tokens: {removed_text}"
)

prompt_embeds = self.text_encoder_3(text_input_ids.to(device))[0]
Expand Down Expand Up @@ -323,6 +324,7 @@ def encode_prompt(
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
clip_skip: Optional[int] = None,
max_sequence_length: Optional[int] = None,
asomoza marked this conversation as resolved.
Show resolved Hide resolved
):
r"""

Expand Down Expand Up @@ -403,6 +405,7 @@ def encode_prompt(
t5_prompt_embed = self._get_t5_prompt_embeds(
prompt=prompt_3,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
device=device,
)

Expand Down Expand Up @@ -456,7 +459,10 @@ def encode_prompt(
negative_clip_prompt_embeds = torch.cat([negative_prompt_embed, negative_prompt_2_embed], dim=-1)

t5_negative_prompt_embed = self._get_t5_prompt_embeds(
prompt=negative_prompt_3, num_images_per_prompt=num_images_per_prompt, device=device
prompt=negative_prompt_3,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
device=device,
)

negative_clip_prompt_embeds = torch.nn.functional.pad(
Expand Down Expand Up @@ -486,6 +492,7 @@ def check_inputs(
pooled_prompt_embeds=None,
negative_pooled_prompt_embeds=None,
callback_on_step_end_tensor_inputs=None,
max_sequence_length=None,
):
if height % 8 != 0 or width % 8 != 0:
raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
Expand Down Expand Up @@ -557,6 +564,9 @@ def check_inputs(
"If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
)

if max_sequence_length is not None and max_sequence_length > 512:
raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")

def prepare_latents(
self,
batch_size,
Expand Down Expand Up @@ -643,6 +653,7 @@ def __call__(
clip_skip: Optional[int] = None,
callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
callback_on_step_end_tensor_inputs: List[str] = ["latents"],
max_sequence_length: Optional[int] = None,
):
r"""
Function invoked when calling the pipeline for generation.
Expand Down Expand Up @@ -726,6 +737,7 @@ def __call__(
The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
`._callback_tensor_inputs` attribute of your pipeline class.
max_sequence_length (`int` defaults to 256): Maximum sequence length to use with the `prompt`.

Examples:

Expand Down Expand Up @@ -753,6 +765,7 @@ def __call__(
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
max_sequence_length=max_sequence_length,
)

self._guidance_scale = guidance_scale
Expand Down Expand Up @@ -790,6 +803,7 @@ def __call__(
device=device,
clip_skip=self.clip_skip,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
)

if self.do_classifier_free_guidance:
Expand Down
22 changes: 18 additions & 4 deletions src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -220,6 +220,7 @@ def _get_t5_prompt_embeds(
self,
prompt: Union[str, List[str]] = None,
num_images_per_prompt: int = 1,
max_sequence_length: int = 256,
device: Optional[torch.device] = None,
dtype: Optional[torch.dtype] = None,
):
Expand All @@ -239,7 +240,7 @@ def _get_t5_prompt_embeds(
text_inputs = self.tokenizer_3(
prompt,
padding="max_length",
max_length=self.tokenizer_max_length,
max_length=max_sequence_length,
truncation=True,
add_special_tokens=True,
return_tensors="pt",
Expand All @@ -250,8 +251,8 @@ def _get_t5_prompt_embeds(
if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
removed_text = self.tokenizer_3.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
logger.warning(
"The following part of your input was truncated because CLIP can only handle sequences up to"
f" {self.tokenizer_max_length} tokens: {removed_text}"
"The following part of your input was truncated because `max_sequence_length` is set to "
f" {max_sequence_length} tokens: {removed_text}"
)

prompt_embeds = self.text_encoder_3(text_input_ids.to(device))[0]
Expand Down Expand Up @@ -340,6 +341,7 @@ def encode_prompt(
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
clip_skip: Optional[int] = None,
max_sequence_length: Optional[int] = None,
asomoza marked this conversation as resolved.
Show resolved Hide resolved
):
r"""

Expand Down Expand Up @@ -420,6 +422,7 @@ def encode_prompt(
t5_prompt_embed = self._get_t5_prompt_embeds(
prompt=prompt_3,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
device=device,
)

Expand Down Expand Up @@ -473,7 +476,10 @@ def encode_prompt(
negative_clip_prompt_embeds = torch.cat([negative_prompt_embed, negative_prompt_2_embed], dim=-1)

t5_negative_prompt_embed = self._get_t5_prompt_embeds(
prompt=negative_prompt_3, num_images_per_prompt=num_images_per_prompt, device=device
prompt=negative_prompt_3,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
device=device,
)

negative_clip_prompt_embeds = torch.nn.functional.pad(
Expand Down Expand Up @@ -502,6 +508,7 @@ def check_inputs(
pooled_prompt_embeds=None,
negative_pooled_prompt_embeds=None,
callback_on_step_end_tensor_inputs=None,
max_sequence_length=None,
):
if strength < 0 or strength > 1:
raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
Expand Down Expand Up @@ -573,6 +580,9 @@ def check_inputs(
"If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
)

if max_sequence_length is not None and max_sequence_length > 512:
raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")

def get_timesteps(self, num_inference_steps, strength, device):
# get the original timestep using init_timestep
init_timestep = min(num_inference_steps * strength, num_inference_steps)
Expand Down Expand Up @@ -684,6 +694,7 @@ def __call__(
clip_skip: Optional[int] = None,
callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
callback_on_step_end_tensor_inputs: List[str] = ["latents"],
max_sequence_length: Optional[int] = None,
asomoza marked this conversation as resolved.
Show resolved Hide resolved
):
r"""
Function invoked when calling the pipeline for generation.
Expand Down Expand Up @@ -763,6 +774,7 @@ def __call__(
The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
`._callback_tensor_inputs` attribute of your pipeline class.
max_sequence_length (`int` defaults to 256): Maximum sequence length to use with the `prompt`.

Examples:

Expand All @@ -786,6 +798,7 @@ def __call__(
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
max_sequence_length=max_sequence_length,
)

self._guidance_scale = guidance_scale
Expand Down Expand Up @@ -822,6 +835,7 @@ def __call__(
device=device,
clip_skip=self.clip_skip,
num_images_per_prompt=num_images_per_prompt,
max_sequence_length=max_sequence_length,
)

if self.do_classifier_free_guidance:
Expand Down
Loading